Filter and Cascade size / hash count should be specifiable in terms of entry count / error rate

[mozmark]

Ideally, the mechanism to set these should match whatever we're using on the backend (e.g. should be easy enough to copy the approach taken by py-bloom)